What the cash drawer reconciler costs
A shop with two tills trading six days a week makes about fifty closes a month; a small chain with five sites makes a few hundred. Nothing here is always-on, and the messaging line is unusually small because the whole design is about not sending messages. Here is where each cent goes.
Key takeaways
- About $3 a month at 180 closes. Roughly $6 at 600 closes.
- One Bedrock read per close is the only line that scales. Everything else is rounding.
- Nothing is always-on, so a quiet month genuinely costs almost nothing.
- The messaging line is tiny on purpose: a good month sends almost no mail at all.
- The duplicate test runs before the read, so resends are free.
- The three real risks: a retry loop, storage nobody expires, and a bigger model than the job needs.
The bill at three volumes
These are US East prices at the time of writing, at three volumes that bracket most small businesses. Find the bar closest to your own and read across.
Line by line
| Line | At 180 closes | How it scales |
|---|---|---|
| Bedrock read | $0.61 | Linear. One call per close, roughly 1,800 in and 200 out tokens. |
| Textract | $0.68 | Linear at $0.0038 per close. |
| SES | $0.10 | Linear. About 0.4 messages per close. |
| DynamoDB + S3 | $0.31 | Storage grows with what you retain, not with throughput. |
| Lambda + SQS | $0.12 | Linear, and effectively free at this scale. |
| CloudWatch | $0.16 | Flat, if you set retention. Unbounded if you do not. |
| Secrets Manager | $0.40 | Flat. One secret, $0.40 a month. |
| AWS Budgets | $0.46 | Flat. Two actions, so you find out before the bill does. |
The Textract line is the biggest per-close cost here, because most closes involve two photographs rather than one — the count sheet and, where the point of sale only prints, the till printout. A POS that emails its export halves it.
The three ways this bill surprises you
Every one of these has happened to somebody, and all three are cheap to prevent.
- A retry loop on an unreadable photo. A dark or angled count sheet makes Textract return nothing useful, the function throws, and the queue redelivers. Without a dead-letter queue that is one bad photo costing more than a month of good ones. A maximum receive count of three fixes it.
- Re-reading on every recount. A supervisor recounting and rephotographing three times pays for three full reads unless the cache is keyed on the image digest rather than the shift id.
- Keeping every photo forever. Count sheets are small but a hundred and twelve a month for seven years is not. An S3 lifecycle rule tiering at ninety days and expiring at your actual record-keeping horizon keeps the storage line flat.
What it costs when nothing happens
This matters more than the headline number for a seasonal business. In a month with nothing to process the bill is the fixed band: Secrets Manager at forty cents, AWS Budgets at forty-six, and a few cents of storage. Call it a dollar. There is no instance to stop and nothing to remember to turn off.
- Management
- Analytics
- Front-end & mobile
Set these on day one
- A dead-letter queue on every SQS queue, with a maximum receive count of three.
- Thirty-day retention on every CloudWatch log group. There is no default that is safe.
- An S3 lifecycle rule on the object prefix, tiering at 90 days and expiring at your actual record-keeping horizon.
- Two AWS Budgets actions — one that emails at half your expected spend, one at double it. The second is how you find out about a loop in an hour instead of a month.
- Provisioned concurrency: none. Nothing here is latency-sensitive enough to justify paying for a warm function.
Next: the same system drawn for engineers — service names, resource identifiers, IAM scopes, table schemas and the model id.
All posts